Direction of arrival estimation for coherent signals based on convolutional neural network with attention mechanism

A direction of arrival (DOA) estimation method for coherent signals is proposed based on an improved convolutional neural network (CNN) with attention mechanism. The real, imaginary and phase components of the covariance matrix are employed as CNN multi-channel inputs. A regression model is adopted as CNN output which avoids the grid mismatch problem caused by multi-classification output. Moreover, to enhance the accuracy of the estimation, attention mechanism is added to the CNN, which could give more weight to the main features effectively. The simulation results demonstrate that the proposed method surpasses the performance of other approaches, which is suitable to resolve coherent signals in multipath environments.